Summary

  • For 2025Q4 the StatBank cube MEC02 prints three rows in gigawatt hours — 8,420 for "All metered electricity consumption", 1,991 for "Data centres" and 6,429 for "Customers other than data centres" — and it prints no percentage row and no per-facility row.
  • The office states that data centres were not readily identifiable in the meter data, that it examined around 2.6 million meters to identify Meter Point Reference Numbers itself, and that "There is no agreed definition of a data centre."

Start with what the office actually put in the MEC02 dataset. One statistic, coded C03907V04659 and labelled "Electricity Consumption", in gigawatt hours, zero decimals. The time dimension runs from 2015Q1 to 2025Q4 — forty-four quarterly periods, the last of them 2025Q4 — and the cube was updated on 7 July 2026.

The classification dimension holds exactly three values, and their labels are worth reading in full: "All metered electricity consumption" for the total, "Data centres" for the category, and "Customers other than data centres" for what remains. For 2025Q4 those three rows read 8,420 GWh, 1,991 GWh and 6,429 GWh. There is no share field anywhere in the cube. There is no row for a named facility, no largest-operator row, no top-ten line.

So every percentage attached to this release is somebody's arithmetic, including mine. 1,991 divided by 8,420 is 23.65% — a figure I computed from two printed rows, not a figure the Central Statistics Office published. That distinction is usually buried in a footnote or lost altogether when the number travels. It should be the first thing a reader sees, because the arithmetic is the easy part and the classification is the hard part.

Now the classification. The Central Statistics Office describes the pipeline in its background notes, and the first sentence fixes the supply chain: "The data were extracted by the CSO from meter data received from the Electricity Supply Board (ESB Networks)." That is the Electricity Supply Board in its metering role — the network reading the dials, not the body deciding what the dials mean.

Because the second sentence is the one that reorganises the whole story: "Data centres were not readily identifiable in the meter data so the CSO examined around 2.6 million meters to identify Meter Point Reference Numbers (MPRNs)." The category does not arrive with the meter reading. It is built afterwards, by the statistical office, out of a sweep of roughly 2.6 million accounts.

The quality report for the 2025 reporting period explains why the sweep had to be that wide. "The data received by the CSO were classified by tariff category (DUoS (Distribution Use of System))," it states, and then: "Data centres were found in different DUoS groups hence the need for the CSO to examine all meters to identify them." The meter carries a tariff class, not a building type. Data centres are scattered across those classes, so no single tariff band can stand in for them. The same report records that the Electricity Supply Board supplied quarterly metered consumption data for 2015 to 2025, which is why the series can be run backwards with confidence but cannot be run forwards without the office repeating the exercise.

The awkward detail sits in the next line of the notes: "These meters are monitored in 15-minute periods and the largest data centres are generally classified by ESB Networks as Large Energy Users." The biggest facilities in the country are, in the source data, filed under a tariff grouping built from several distribution-use-of-system bands rather than under anything resembling a data-centre label. Whatever the published row calls them, the network does not.

What the office did instead is a set of inferences, and it lists them: examining the business activity of the Large Energy Users; name and alias searches above 0.5 GWh; examination of business parks above 0.5 GWh; every meter above 1 GWh; other reports; the CSO Business Register; internet searches. Then the sentence that reads like a description of the residual category rather than an admission: "The CSO identified MPRNs, with a high metered electricity consumption, that were not associated with high employment nor with other activity such as cement manufacture." High load, few staff, not a cement works.

That is a reasonable operational test and a very different object from a definition.

The office says so itself, in six words that no headline has ever carried: "There is no agreed definition of a data centre." Read that beside the 2.6 million meters and the three printed rows and the structure of the number becomes clear. The denominator is every meter in the State. The numerator is a set of meter points the statistical office decided to call data centres, on evidence about activity, names and load. Same institution, both sides of the fraction, one published aggregate.

That aggregate is a statement about Irish metered consumption. It is not a statement about any centre in it. No facility is named by MEC02, no facility can point to its own row, and no facility can contest its inclusion through the statistic, because the statistic never reports the decision at that level. A planning submission that borrows the state-level ratio to argue about one site is borrowing a number that was never about a site. The arithmetic survives the translation; the comparison does not.

One limit belongs here rather than in a closing paragraph. The rendered release pages on the office's main site did not return readable text to this desk on 18 November 2026; the machine-readable dataset, the background notes and the quality report did. That is a retrieval limit on our side. It is not evidence that the office withholds its method, and the method documents it does publish are unusually explicit about exactly how uncertain the category is.

Set that against the German efficiency statute examined earlier this week, where the law wrote the definition and the operator held the meter, so the public-body certification rung landed precisely on the definitional floor. Ireland moves the ruler: here the office writes the category and the total, the network supplies the reading and the tariff class, and the operator contributes nothing to the number but consumption. The document a sceptical reader should audit is therefore not the headline and not even the meter. It is the classification note.

Our own Irish coverage has so far read permission and coverage instruments — connection queues with their capacity bands and milestones, incident-reporting clocks, reach and multiplex denominators — every one of them answering what was applied for or granted. This is the first time the archive has read the measured result instead of the permission.

Ireland's data-centre share is not a measurement that fell out of a meter. It is an identification made by an office, after the fact, with a stated method and a stated absence of definition. Quote the rows, label the division, and the argument gets better rather than weaker. Quote the percentage as if the office had printed it, and you are defending a number whose author you cannot name.